drivers/lightnvm/pblk-core.c

   1 /*
   2  * Copyright (C) 2016 CNEX Labs
   3  * Initial release: Javier Gonzalez <javier@cnexlabs.com>
   4  *                  Matias Bjorling <matias@cnexlabs.com>
   5  *
   6  * This program is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU General Public License version
   8  * 2 as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License for more details.
  14  *
  15  * pblk-core.c - pblk's core functionality
  16  *
  17  */
  18
  19 #include "pblk.h"
  20 #include <linux/time.h>
  21
  22 static void pblk_mark_bb(struct pblk *pblk, struct pblk_line *line,
  23                          struct ppa_addr *ppa)
  24 {
  25         struct nvm_tgt_dev *dev = pblk->dev;
  26         struct nvm_geo *geo = &dev->geo;
  27         int pos = pblk_dev_ppa_to_pos(geo, *ppa);
  28
  29         pr_debug("pblk: erase failed: line:%d, pos:%d\n", line->id, pos);
  30         atomic_long_inc(&pblk->erase_failed);
  31
  32         atomic_dec(&line->blk_in_line);
  33         if (test_and_set_bit(pos, line->blk_bitmap))
  34                 pr_err("pblk: attempted to erase bb: line:%d, pos:%d\n",
  35                                                         line->id, pos);
  36
  37         pblk_line_run_ws(pblk, NULL, ppa, pblk_line_mark_bb);
  38 }
  39
  40 static void __pblk_end_io_erase(struct pblk *pblk, struct nvm_rq *rqd)
  41 {
  42         struct pblk_line *line;
  43
  44         line = &pblk->lines[pblk_dev_ppa_to_line(rqd->ppa_addr)];
  45         atomic_dec(&line->left_seblks);
  46
  47         if (rqd->error) {
  48                 struct ppa_addr *ppa;
  49
  50                 ppa = kmalloc(sizeof(struct ppa_addr), GFP_ATOMIC);
  51                 if (!ppa)
  52                         return;
  53
  54                 *ppa = rqd->ppa_addr;
  55                 pblk_mark_bb(pblk, line, ppa);
  56         }
  57 }
  58
  59 /* Erase completion assumes that only one block is erased at the time */
  60 static void pblk_end_io_erase(struct nvm_rq *rqd)
  61 {
  62         struct pblk *pblk = rqd->private;
  63
  64         up(&pblk->erase_sem);
  65         __pblk_end_io_erase(pblk, rqd);
  66         mempool_free(rqd, pblk->r_rq_pool);
  67 }
  68
  69 static void __pblk_map_invalidate(struct pblk *pblk, struct pblk_line *line,
  70                                   u64 paddr)
  71 {
  72         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
  73         struct list_head *move_list = NULL;
  74
  75         /* Lines being reclaimed (GC'ed) cannot be invalidated. Before the L2P
  76          * table is modified with reclaimed sectors, a check is done to endure
  77          * that newer updates are not overwritten.
  78          */
  79         spin_lock(&line->lock);
  80         if (line->state == PBLK_LINESTATE_GC ||
  81                                         line->state == PBLK_LINESTATE_FREE) {
  82                 spin_unlock(&line->lock);
  83                 return;
  84         }
  85
  86         if (test_and_set_bit(paddr, line->invalid_bitmap)) {
  87                 WARN_ONCE(1, "pblk: double invalidate\n");
  88                 spin_unlock(&line->lock);
  89                 return;
  90         }
  91         line->vsc--;
  92
  93         if (line->state == PBLK_LINESTATE_CLOSED)
  94                 move_list = pblk_line_gc_list(pblk, line);
  95         spin_unlock(&line->lock);
  96
  97         if (move_list) {
  98                 spin_lock(&l_mg->gc_lock);
  99                 spin_lock(&line->lock);
 100                 /* Prevent moving a line that has just been chosen for GC */
 101                 if (line->state == PBLK_LINESTATE_GC ||
 102                                         line->state == PBLK_LINESTATE_FREE) {
 103                         spin_unlock(&line->lock);
 104                         spin_unlock(&l_mg->gc_lock);
 105                         return;
 106                 }
 107                 spin_unlock(&line->lock);
 108
 109                 list_move_tail(&line->list, move_list);
 110                 spin_unlock(&l_mg->gc_lock);
 111         }
 112 }
 113
 114 void pblk_map_invalidate(struct pblk *pblk, struct ppa_addr ppa)
 115 {
 116         struct pblk_line *line;
 117         u64 paddr;
 118         int line_id;
 119
 120 #ifdef CONFIG_NVM_DEBUG
 121         /* Callers must ensure that the ppa points to a device address */
 122         BUG_ON(pblk_addr_in_cache(ppa));
 123         BUG_ON(pblk_ppa_empty(ppa));
 124 #endif
 125
 126         line_id = pblk_tgt_ppa_to_line(ppa);
 127         line = &pblk->lines[line_id];
 128         paddr = pblk_dev_ppa_to_line_addr(pblk, ppa);
 129
 130         __pblk_map_invalidate(pblk, line, paddr);
 131 }
 132
 133 void pblk_map_pad_invalidate(struct pblk *pblk, struct pblk_line *line,
 134                              u64 paddr)
 135 {
 136         __pblk_map_invalidate(pblk, line, paddr);
 137
 138         pblk_rb_sync_init(&pblk->rwb, NULL);
 139         line->left_ssecs--;
 140         if (!line->left_ssecs)
 141                 pblk_line_run_ws(pblk, line, NULL, pblk_line_close_ws);
 142         pblk_rb_sync_end(&pblk->rwb, NULL);
 143 }
 144
 145 static void pblk_invalidate_range(struct pblk *pblk, sector_t slba,
 146                                   unsigned int nr_secs)
 147 {
 148         sector_t lba;
 149
 150         spin_lock(&pblk->trans_lock);
 151         for (lba = slba; lba < slba + nr_secs; lba++) {
 152                 struct ppa_addr ppa;
 153
 154                 ppa = pblk_trans_map_get(pblk, lba);
 155
 156                 if (!pblk_addr_in_cache(ppa) && !pblk_ppa_empty(ppa))
 157                         pblk_map_invalidate(pblk, ppa);
 158
 159                 pblk_ppa_set_empty(&ppa);
 160                 pblk_trans_map_set(pblk, lba, ppa);
 161         }
 162         spin_unlock(&pblk->trans_lock);
 163 }
 164
 165 struct nvm_rq *pblk_alloc_rqd(struct pblk *pblk, int rw)
 166 {
 167         mempool_t *pool;
 168         struct nvm_rq *rqd;
 169         int rq_size;
 170
 171         if (rw == WRITE) {
 172                 pool = pblk->w_rq_pool;
 173                 rq_size = pblk_w_rq_size;
 174         } else {
 175                 pool = pblk->r_rq_pool;
 176                 rq_size = pblk_r_rq_size;
 177         }
 178
 179         rqd = mempool_alloc(pool, GFP_KERNEL);
 180         memset(rqd, 0, rq_size);
 181
 182         return rqd;
 183 }
 184
 185 void pblk_free_rqd(struct pblk *pblk, struct nvm_rq *rqd, int rw)
 186 {
 187         mempool_t *pool;
 188
 189         if (rw == WRITE)
 190                 pool = pblk->w_rq_pool;
 191         else
 192                 pool = pblk->r_rq_pool;
 193
 194         mempool_free(rqd, pool);
 195 }
 196
 197 void pblk_bio_free_pages(struct pblk *pblk, struct bio *bio, int off,
 198                          int nr_pages)
 199 {
 200         struct bio_vec bv;
 201         int i;
 202
 203         WARN_ON(off + nr_pages != bio->bi_vcnt);
 204
 205         bio_advance(bio, off * PBLK_EXPOSED_PAGE_SIZE);
 206         for (i = off; i < nr_pages + off; i++) {
 207                 bv = bio->bi_io_vec[i];
 208                 mempool_free(bv.bv_page, pblk->page_pool);
 209         }
 210 }
 211
 212 int pblk_bio_add_pages(struct pblk *pblk, struct bio *bio, gfp_t flags,
 213                        int nr_pages)
 214 {
 215         struct request_queue *q = pblk->dev->q;
 216         struct page *page;
 217         int i, ret;
 218
 219         for (i = 0; i < nr_pages; i++) {
 220                 page = mempool_alloc(pblk->page_pool, flags);
 221                 if (!page)
 222                         goto err;
 223
 224                 ret = bio_add_pc_page(q, bio, page, PBLK_EXPOSED_PAGE_SIZE, 0);
 225                 if (ret != PBLK_EXPOSED_PAGE_SIZE) {
 226                         pr_err("pblk: could not add page to bio\n");
 227                         mempool_free(page, pblk->page_pool);
 228                         goto err;
 229                 }
 230         }
 231
 232         return 0;
 233 err:
 234         pblk_bio_free_pages(pblk, bio, 0, i - 1);
 235         return -1;
 236 }
 237
 238 static void pblk_write_kick(struct pblk *pblk)
 239 {
 240         wake_up_process(pblk->writer_ts);
 241         mod_timer(&pblk->wtimer, jiffies + msecs_to_jiffies(1000));
 242 }
 243
 244 void pblk_write_timer_fn(unsigned long data)
 245 {
 246         struct pblk *pblk = (struct pblk *)data;
 247
 248         /* kick the write thread every tick to flush outstanding data */
 249         pblk_write_kick(pblk);
 250 }
 251
 252 void pblk_write_should_kick(struct pblk *pblk)
 253 {
 254         unsigned int secs_avail = pblk_rb_read_count(&pblk->rwb);
 255
 256         if (secs_avail >= pblk->min_write_pgs)
 257                 pblk_write_kick(pblk);
 258 }
 259
 260 void pblk_end_bio_sync(struct bio *bio)
 261 {
 262         struct completion *waiting = bio->bi_private;
 263
 264         complete(waiting);
 265 }
 266
 267 void pblk_end_io_sync(struct nvm_rq *rqd)
 268 {
 269         struct completion *waiting = rqd->private;
 270
 271         complete(waiting);
 272 }
 273
 274 void pblk_flush_writer(struct pblk *pblk)
 275 {
 276         struct bio *bio;
 277         int ret;
 278         DECLARE_COMPLETION_ONSTACK(wait);
 279
 280         bio = bio_alloc(GFP_KERNEL, 1);
 281         if (!bio)
 282                 return;
 283
 284         bio->bi_iter.bi_sector = 0; /* internal bio */
 285         bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_OP_FLUSH);
 286         bio->bi_private = &wait;
 287         bio->bi_end_io = pblk_end_bio_sync;
 288
 289         ret = pblk_write_to_cache(pblk, bio, 0);
 290         if (ret == NVM_IO_OK) {
 291                 if (!wait_for_completion_io_timeout(&wait,
 292                                 msecs_to_jiffies(PBLK_COMMAND_TIMEOUT_MS))) {
 293                         pr_err("pblk: flush cache timed out\n");
 294                 }
 295         } else if (ret != NVM_IO_DONE) {
 296                 pr_err("pblk: tear down bio failed\n");
 297         }
 298
 299         if (bio->bi_error)
 300                 pr_err("pblk: flush sync write failed (%u)\n", bio->bi_error);
 301
 302         bio_put(bio);
 303 }
 304
 305 struct list_head *pblk_line_gc_list(struct pblk *pblk, struct pblk_line *line)
 306 {
 307         struct pblk_line_meta *lm = &pblk->lm;
 308         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 309         struct list_head *move_list = NULL;
 310
 311         if (!line->vsc) {
 312                 if (line->gc_group != PBLK_LINEGC_FULL) {
 313                         line->gc_group = PBLK_LINEGC_FULL;
 314                         move_list = &l_mg->gc_full_list;
 315                 }
 316         } else if (line->vsc < lm->mid_thrs) {
 317                 if (line->gc_group != PBLK_LINEGC_HIGH) {
 318                         line->gc_group = PBLK_LINEGC_HIGH;
 319                         move_list = &l_mg->gc_high_list;
 320                 }
 321         } else if (line->vsc < lm->high_thrs) {
 322                 if (line->gc_group != PBLK_LINEGC_MID) {
 323                         line->gc_group = PBLK_LINEGC_MID;
 324                         move_list = &l_mg->gc_mid_list;
 325                 }
 326         } else if (line->vsc < line->sec_in_line) {
 327                 if (line->gc_group != PBLK_LINEGC_LOW) {
 328                         line->gc_group = PBLK_LINEGC_LOW;
 329                         move_list = &l_mg->gc_low_list;
 330                 }
 331         } else if (line->vsc == line->sec_in_line) {
 332                 if (line->gc_group != PBLK_LINEGC_EMPTY) {
 333                         line->gc_group = PBLK_LINEGC_EMPTY;
 334                         move_list = &l_mg->gc_empty_list;
 335                 }
 336         } else {
 337                 line->state = PBLK_LINESTATE_CORRUPT;
 338                 line->gc_group = PBLK_LINEGC_NONE;
 339                 move_list =  &l_mg->corrupt_list;
 340                 pr_err("pblk: corrupted vsc for line %d, vsc:%d (%d/%d/%d)\n",
 341                                                 line->id, line->vsc,
 342                                                 line->sec_in_line,
 343                                                 lm->high_thrs, lm->mid_thrs);
 344         }
 345
 346         return move_list;
 347 }
 348
 349 void pblk_discard(struct pblk *pblk, struct bio *bio)
 350 {
 351         sector_t slba = pblk_get_lba(bio);
 352         sector_t nr_secs = pblk_get_secs(bio);
 353
 354         pblk_invalidate_range(pblk, slba, nr_secs);
 355 }
 356
 357 struct ppa_addr pblk_get_lba_map(struct pblk *pblk, sector_t lba)
 358 {
 359         struct ppa_addr ppa;
 360
 361         spin_lock(&pblk->trans_lock);
 362         ppa = pblk_trans_map_get(pblk, lba);
 363         spin_unlock(&pblk->trans_lock);
 364
 365         return ppa;
 366 }
 367
 368 void pblk_log_write_err(struct pblk *pblk, struct nvm_rq *rqd)
 369 {
 370         atomic_long_inc(&pblk->write_failed);
 371 #ifdef CONFIG_NVM_DEBUG
 372         pblk_print_failed_rqd(pblk, rqd, rqd->error);
 373 #endif
 374 }
 375
 376 void pblk_log_read_err(struct pblk *pblk, struct nvm_rq *rqd)
 377 {
 378         /* Empty page read is not necessarily an error (e.g., L2P recovery) */
 379         if (rqd->error == NVM_RSP_ERR_EMPTYPAGE) {
 380                 atomic_long_inc(&pblk->read_empty);
 381                 return;
 382         }
 383
 384         switch (rqd->error) {
 385         case NVM_RSP_WARN_HIGHECC:
 386                 atomic_long_inc(&pblk->read_high_ecc);
 387                 break;
 388         case NVM_RSP_ERR_FAILECC:
 389         case NVM_RSP_ERR_FAILCRC:
 390                 atomic_long_inc(&pblk->read_failed);
 391                 break;
 392         default:
 393                 pr_err("pblk: unknown read error:%d\n", rqd->error);
 394         }
 395 #ifdef CONFIG_NVM_DEBUG
 396         pblk_print_failed_rqd(pblk, rqd, rqd->error);
 397 #endif
 398 }
 399
 400 int pblk_submit_io(struct pblk *pblk, struct nvm_rq *rqd)
 401 {
 402         struct nvm_tgt_dev *dev = pblk->dev;
 403
 404 #ifdef CONFIG_NVM_DEBUG
 405         struct ppa_addr *ppa_list;
 406
 407         ppa_list = (rqd->nr_ppas > 1) ? rqd->ppa_list : &rqd->ppa_addr;
 408         if (pblk_boundary_ppa_checks(dev, ppa_list, rqd->nr_ppas)) {
 409                 WARN_ON(1);
 410                 return -EINVAL;
 411         }
 412
 413         if (rqd->opcode == NVM_OP_PWRITE) {
 414                 struct pblk_line *line;
 415                 struct ppa_addr ppa;
 416                 int i;
 417
 418                 for (i = 0; i < rqd->nr_ppas; i++) {
 419                         ppa = ppa_list[i];
 420                         line = &pblk->lines[pblk_dev_ppa_to_line(ppa)];
 421
 422                         spin_lock(&line->lock);
 423                         if (line->state != PBLK_LINESTATE_OPEN) {
 424                                 pr_err("pblk: bad ppa: line:%d,state:%d\n",
 425                                                         line->id, line->state);
 426                                 WARN_ON(1);
 427                                 spin_unlock(&line->lock);
 428                                 return -EINVAL;
 429                         }
 430                         spin_unlock(&line->lock);
 431                 }
 432         }
 433 #endif
 434         return nvm_submit_io(dev, rqd);
 435 }
 436
 437 struct bio *pblk_bio_map_addr(struct pblk *pblk, void *data,
 438                               unsigned int nr_secs, unsigned int len,
 439                               gfp_t gfp_mask)
 440 {
 441         struct nvm_tgt_dev *dev = pblk->dev;
 442         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 443         void *kaddr = data;
 444         struct page *page;
 445         struct bio *bio;
 446         int i, ret;
 447
 448         if (l_mg->emeta_alloc_type == PBLK_KMALLOC_META)
 449                 return bio_map_kern(dev->q, kaddr, len, gfp_mask);
 450
 451         bio = bio_kmalloc(gfp_mask, nr_secs);
 452         if (!bio)
 453                 return ERR_PTR(-ENOMEM);
 454
 455         for (i = 0; i < nr_secs; i++) {
 456                 page = vmalloc_to_page(kaddr);
 457                 if (!page) {
 458                         pr_err("pblk: could not map vmalloc bio\n");
 459                         bio_put(bio);
 460                         bio = ERR_PTR(-ENOMEM);
 461                         goto out;
 462                 }
 463
 464                 ret = bio_add_pc_page(dev->q, bio, page, PAGE_SIZE, 0);
 465                 if (ret != PAGE_SIZE) {
 466                         pr_err("pblk: could not add page to bio\n");
 467                         bio_put(bio);
 468                         bio = ERR_PTR(-ENOMEM);
 469                         goto out;
 470                 }
 471
 472                 kaddr += PAGE_SIZE;
 473         }
 474 out:
 475         return bio;
 476 }
 477
 478 int pblk_calc_secs(struct pblk *pblk, unsigned long secs_avail,
 479                    unsigned long secs_to_flush)
 480 {
 481         int max = pblk->max_write_pgs;
 482         int min = pblk->min_write_pgs;
 483         int secs_to_sync = 0;
 484
 485         if (secs_avail >= max)
 486                 secs_to_sync = max;
 487         else if (secs_avail >= min)
 488                 secs_to_sync = min * (secs_avail / min);
 489         else if (secs_to_flush)
 490                 secs_to_sync = min;
 491
 492         return secs_to_sync;
 493 }
 494
 495 static u64 __pblk_alloc_page(struct pblk *pblk, struct pblk_line *line,
 496                              int nr_secs)
 497 {
 498         u64 addr;
 499         int i;
 500
 501         /* logic error: ppa out-of-bounds. Prevent generating bad address */
 502         if (line->cur_sec + nr_secs > pblk->lm.sec_per_line) {
 503                 WARN(1, "pblk: page allocation out of bounds\n");
 504                 nr_secs = pblk->lm.sec_per_line - line->cur_sec;
 505         }
 506
 507         line->cur_sec = addr = find_next_zero_bit(line->map_bitmap,
 508                                         pblk->lm.sec_per_line, line->cur_sec);
 509         for (i = 0; i < nr_secs; i++, line->cur_sec++)
 510                 WARN_ON(test_and_set_bit(line->cur_sec, line->map_bitmap));
 511
 512         return addr;
 513 }
 514
 515 u64 pblk_alloc_page(struct pblk *pblk, struct pblk_line *line, int nr_secs)
 516 {
 517         u64 addr;
 518
 519         /* Lock needed in case a write fails and a recovery needs to remap
 520          * failed write buffer entries
 521          */
 522         spin_lock(&line->lock);
 523         addr = __pblk_alloc_page(pblk, line, nr_secs);
 524         line->left_msecs -= nr_secs;
 525         WARN(line->left_msecs < 0, "pblk: page allocation out of bounds\n");
 526         spin_unlock(&line->lock);
 527
 528         return addr;
 529 }
 530
 531 /*
 532  * Submit emeta to one LUN in the raid line at the time to avoid a deadlock when
 533  * taking the per LUN semaphore.
 534  */
 535 static int pblk_line_submit_emeta_io(struct pblk *pblk, struct pblk_line *line,
 536                                      u64 paddr, int dir)
 537 {
 538         struct nvm_tgt_dev *dev = pblk->dev;
 539         struct nvm_geo *geo = &dev->geo;
 540         struct pblk_line_meta *lm = &pblk->lm;
 541         struct bio *bio;
 542         struct nvm_rq rqd;
 543         struct ppa_addr *ppa_list;
 544         dma_addr_t dma_ppa_list;
 545         void *emeta = line->emeta;
 546         int min = pblk->min_write_pgs;
 547         int left_ppas = lm->emeta_sec;
 548         int id = line->id;
 549         int rq_ppas, rq_len;
 550         int cmd_op, bio_op;
 551         int flags;
 552         int i, j;
 553         int ret;
 554         DECLARE_COMPLETION_ONSTACK(wait);
 555
 556         if (dir == WRITE) {
 557                 bio_op = REQ_OP_WRITE;
 558                 cmd_op = NVM_OP_PWRITE;
 559                 flags = pblk_set_progr_mode(pblk, WRITE);
 560         } else if (dir == READ) {
 561                 bio_op = REQ_OP_READ;
 562                 cmd_op = NVM_OP_PREAD;
 563                 flags = pblk_set_read_mode(pblk);
 564         } else
 565                 return -EINVAL;
 566
 567         ppa_list = nvm_dev_dma_alloc(dev->parent, GFP_KERNEL, &dma_ppa_list);
 568         if (!ppa_list)
 569                 return -ENOMEM;
 570
 571 next_rq:
 572         memset(&rqd, 0, sizeof(struct nvm_rq));
 573
 574         rq_ppas = pblk_calc_secs(pblk, left_ppas, 0);
 575         rq_len = rq_ppas * geo->sec_size;
 576
 577         bio = pblk_bio_map_addr(pblk, emeta, rq_ppas, rq_len, GFP_KERNEL);
 578         if (IS_ERR(bio)) {
 579                 ret = PTR_ERR(bio);
 580                 goto free_rqd_dma;
 581         }
 582
 583         bio->bi_iter.bi_sector = 0; /* internal bio */
 584         bio_set_op_attrs(bio, bio_op, 0);
 585
 586         rqd.bio = bio;
 587         rqd.opcode = cmd_op;
 588         rqd.flags = flags;
 589         rqd.nr_ppas = rq_ppas;
 590         rqd.ppa_list = ppa_list;
 591         rqd.dma_ppa_list = dma_ppa_list;
 592         rqd.end_io = pblk_end_io_sync;
 593         rqd.private = &wait;
 594
 595         if (dir == WRITE) {
 596                 for (i = 0; i < rqd.nr_ppas; ) {
 597                         spin_lock(&line->lock);
 598                         paddr = __pblk_alloc_page(pblk, line, min);
 599                         spin_unlock(&line->lock);
 600                         for (j = 0; j < min; j++, i++, paddr++)
 601                                 rqd.ppa_list[i] =
 602                                         addr_to_gen_ppa(pblk, paddr, id);
 603                 }
 604         } else {
 605                 for (i = 0; i < rqd.nr_ppas; ) {
 606                         struct ppa_addr ppa = addr_to_gen_ppa(pblk, paddr, id);
 607                         int pos = pblk_dev_ppa_to_pos(geo, ppa);
 608
 609                         while (test_bit(pos, line->blk_bitmap)) {
 610                                 paddr += min;
 611                                 if (pblk_boundary_paddr_checks(pblk, paddr)) {
 612                                         pr_err("pblk: corrupt emeta line:%d\n",
 613                                                                 line->id);
 614                                         bio_put(bio);
 615                                         ret = -EINTR;
 616                                         goto free_rqd_dma;
 617                                 }
 618
 619                                 ppa = addr_to_gen_ppa(pblk, paddr, id);
 620                                 pos = pblk_dev_ppa_to_pos(geo, ppa);
 621                         }
 622
 623                         if (pblk_boundary_paddr_checks(pblk, paddr + min)) {
 624                                 pr_err("pblk: corrupt emeta line:%d\n",
 625                                                                 line->id);
 626                                 bio_put(bio);
 627                                 ret = -EINTR;
 628                                 goto free_rqd_dma;
 629                         }
 630
 631                         for (j = 0; j < min; j++, i++, paddr++)
 632                                 rqd.ppa_list[i] =
 633                                         addr_to_gen_ppa(pblk, paddr, line->id);
 634                 }
 635         }
 636
 637         ret = pblk_submit_io(pblk, &rqd);
 638         if (ret) {
 639                 pr_err("pblk: emeta I/O submission failed: %d\n", ret);
 640                 bio_put(bio);
 641                 goto free_rqd_dma;
 642         }
 643
 644         if (!wait_for_completion_io_timeout(&wait,
 645                                 msecs_to_jiffies(PBLK_COMMAND_TIMEOUT_MS))) {
 646                 pr_err("pblk: emeta I/O timed out\n");
 647         }
 648         reinit_completion(&wait);
 649
 650         bio_put(bio);
 651
 652         if (rqd.error) {
 653                 if (dir == WRITE)
 654                         pblk_log_write_err(pblk, &rqd);
 655                 else
 656                         pblk_log_read_err(pblk, &rqd);
 657         }
 658
 659         emeta += rq_len;
 660         left_ppas -= rq_ppas;
 661         if (left_ppas)
 662                 goto next_rq;
 663 free_rqd_dma:
 664         nvm_dev_dma_free(dev->parent, ppa_list, dma_ppa_list);
 665         return ret;
 666 }
 667
 668 u64 pblk_line_smeta_start(struct pblk *pblk, struct pblk_line *line)
 669 {
 670         struct nvm_tgt_dev *dev = pblk->dev;
 671         struct nvm_geo *geo = &dev->geo;
 672         struct pblk_line_meta *lm = &pblk->lm;
 673         int bit;
 674
 675         /* This usually only happens on bad lines */
 676         bit = find_first_zero_bit(line->blk_bitmap, lm->blk_per_line);
 677         if (bit >= lm->blk_per_line)
 678                 return -1;
 679
 680         return bit * geo->sec_per_pl;
 681 }
 682
 683 static int pblk_line_submit_smeta_io(struct pblk *pblk, struct pblk_line *line,
 684                                      u64 paddr, int dir)
 685 {
 686         struct nvm_tgt_dev *dev = pblk->dev;
 687         struct pblk_line_meta *lm = &pblk->lm;
 688         struct bio *bio;
 689         struct nvm_rq rqd;
 690         __le64 *lba_list = NULL;
 691         int i, ret;
 692         int cmd_op, bio_op;
 693         int flags;
 694         DECLARE_COMPLETION_ONSTACK(wait);
 695
 696         if (dir == WRITE) {
 697                 bio_op = REQ_OP_WRITE;
 698                 cmd_op = NVM_OP_PWRITE;
 699                 flags = pblk_set_progr_mode(pblk, WRITE);
 700                 lba_list = pblk_line_emeta_to_lbas(line->emeta);
 701         } else if (dir == READ) {
 702                 bio_op = REQ_OP_READ;
 703                 cmd_op = NVM_OP_PREAD;
 704                 flags = pblk_set_read_mode(pblk);
 705         } else
 706                 return -EINVAL;
 707
 708         memset(&rqd, 0, sizeof(struct nvm_rq));
 709
 710         rqd.ppa_list = nvm_dev_dma_alloc(dev->parent, GFP_KERNEL,
 711                                                         &rqd.dma_ppa_list);
 712         if (!rqd.ppa_list)
 713                 return -ENOMEM;
 714
 715         bio = bio_map_kern(dev->q, line->smeta, lm->smeta_len, GFP_KERNEL);
 716         if (IS_ERR(bio)) {
 717                 ret = PTR_ERR(bio);
 718                 goto free_ppa_list;
 719         }
 720
 721         bio->bi_iter.bi_sector = 0; /* internal bio */
 722         bio_set_op_attrs(bio, bio_op, 0);
 723
 724         rqd.bio = bio;
 725         rqd.opcode = cmd_op;
 726         rqd.flags = flags;
 727         rqd.nr_ppas = lm->smeta_sec;
 728         rqd.end_io = pblk_end_io_sync;
 729         rqd.private = &wait;
 730
 731         for (i = 0; i < lm->smeta_sec; i++, paddr++) {
 732                 rqd.ppa_list[i] = addr_to_gen_ppa(pblk, paddr, line->id);
 733                 if (dir == WRITE)
 734                         lba_list[paddr] = cpu_to_le64(ADDR_EMPTY);
 735         }
 736
 737         /*
 738          * This I/O is sent by the write thread when a line is replace. Since
 739          * the write thread is the only one sending write and erase commands,
 740          * there is no need to take the LUN semaphore.
 741          */
 742         ret = pblk_submit_io(pblk, &rqd);
 743         if (ret) {
 744                 pr_err("pblk: smeta I/O submission failed: %d\n", ret);
 745                 bio_put(bio);
 746                 goto free_ppa_list;
 747         }
 748
 749         if (!wait_for_completion_io_timeout(&wait,
 750                                 msecs_to_jiffies(PBLK_COMMAND_TIMEOUT_MS))) {
 751                 pr_err("pblk: smeta I/O timed out\n");
 752         }
 753
 754         if (rqd.error) {
 755                 if (dir == WRITE)
 756                         pblk_log_write_err(pblk, &rqd);
 757                 else
 758                         pblk_log_read_err(pblk, &rqd);
 759         }
 760
 761 free_ppa_list:
 762         nvm_dev_dma_free(dev->parent, rqd.ppa_list, rqd.dma_ppa_list);
 763
 764         return ret;
 765 }
 766
 767 int pblk_line_read_smeta(struct pblk *pblk, struct pblk_line *line)
 768 {
 769         u64 bpaddr = pblk_line_smeta_start(pblk, line);
 770
 771         return pblk_line_submit_smeta_io(pblk, line, bpaddr, READ);
 772 }
 773
 774 int pblk_line_read_emeta(struct pblk *pblk, struct pblk_line *line)
 775 {
 776         return pblk_line_submit_emeta_io(pblk, line, line->emeta_ssec, READ);
 777 }
 778
 779 static void pblk_setup_e_rq(struct pblk *pblk, struct nvm_rq *rqd,
 780                             struct ppa_addr ppa)
 781 {
 782         rqd->opcode = NVM_OP_ERASE;
 783         rqd->ppa_addr = ppa;
 784         rqd->nr_ppas = 1;
 785         rqd->flags = pblk_set_progr_mode(pblk, ERASE);
 786         rqd->bio = NULL;
 787 }
 788
 789 static int pblk_blk_erase_sync(struct pblk *pblk, struct ppa_addr ppa)
 790 {
 791         struct nvm_rq rqd;
 792         int ret;
 793         DECLARE_COMPLETION_ONSTACK(wait);
 794
 795         memset(&rqd, 0, sizeof(struct nvm_rq));
 796
 797         pblk_setup_e_rq(pblk, &rqd, ppa);
 798
 799         rqd.end_io = pblk_end_io_sync;
 800         rqd.private = &wait;
 801
 802         /* The write thread schedules erases so that it minimizes disturbances
 803          * with writes. Thus, there is no need to take the LUN semaphore.
 804          */
 805         ret = pblk_submit_io(pblk, &rqd);
 806         if (ret) {
 807                 struct nvm_tgt_dev *dev = pblk->dev;
 808                 struct nvm_geo *geo = &dev->geo;
 809
 810                 pr_err("pblk: could not sync erase line:%d,blk:%d\n",
 811                                         pblk_dev_ppa_to_line(ppa),
 812                                         pblk_dev_ppa_to_pos(geo, ppa));
 813
 814                 rqd.error = ret;
 815                 goto out;
 816         }
 817
 818         if (!wait_for_completion_io_timeout(&wait,
 819                                 msecs_to_jiffies(PBLK_COMMAND_TIMEOUT_MS))) {
 820                 pr_err("pblk: sync erase timed out\n");
 821         }
 822
 823 out:
 824         rqd.private = pblk;
 825         __pblk_end_io_erase(pblk, &rqd);
 826
 827         return 0;
 828 }
 829
 830 int pblk_line_erase(struct pblk *pblk, struct pblk_line *line)
 831 {
 832         struct pblk_line_meta *lm = &pblk->lm;
 833         struct ppa_addr ppa;
 834         int bit = -1;
 835
 836         /* Erase only good blocks, one at a time */
 837         do {
 838                 spin_lock(&line->lock);
 839                 bit = find_next_zero_bit(line->erase_bitmap, lm->blk_per_line,
 840                                                                 bit + 1);
 841                 if (bit >= lm->blk_per_line) {
 842                         spin_unlock(&line->lock);
 843                         break;
 844                 }
 845
 846                 ppa = pblk->luns[bit].bppa; /* set ch and lun */
 847                 ppa.g.blk = line->id;
 848
 849                 atomic_dec(&line->left_eblks);
 850                 WARN_ON(test_and_set_bit(bit, line->erase_bitmap));
 851                 spin_unlock(&line->lock);
 852
 853                 if (pblk_blk_erase_sync(pblk, ppa)) {
 854                         pr_err("pblk: failed to erase line %d\n", line->id);
 855                         return -ENOMEM;
 856                 }
 857         } while (1);
 858
 859         return 0;
 860 }
 861
 862 /* For now lines are always assumed full lines. Thus, smeta former and current
 863  * lun bitmaps are omitted.
 864  */
 865 static int pblk_line_set_metadata(struct pblk *pblk, struct pblk_line *line,
 866                                   struct pblk_line *cur)
 867 {
 868         struct nvm_tgt_dev *dev = pblk->dev;
 869         struct nvm_geo *geo = &dev->geo;
 870         struct pblk_line_meta *lm = &pblk->lm;
 871         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 872         struct line_smeta *smeta = line->smeta;
 873         struct line_emeta *emeta = line->emeta;
 874         int nr_blk_line;
 875
 876         /* After erasing the line, new bad blocks might appear and we risk
 877          * having an invalid line
 878          */
 879         nr_blk_line = lm->blk_per_line -
 880                         bitmap_weight(line->blk_bitmap, lm->blk_per_line);
 881         if (nr_blk_line < lm->min_blk_line) {
 882                 spin_lock(&l_mg->free_lock);
 883                 spin_lock(&line->lock);
 884                 line->state = PBLK_LINESTATE_BAD;
 885                 spin_unlock(&line->lock);
 886
 887                 list_add_tail(&line->list, &l_mg->bad_list);
 888                 spin_unlock(&l_mg->free_lock);
 889
 890                 pr_debug("pblk: line %d is bad\n", line->id);
 891
 892                 return 0;
 893         }
 894
 895         /* Run-time metadata */
 896         line->lun_bitmap = ((void *)(smeta)) + sizeof(struct line_smeta);
 897
 898         /* Mark LUNs allocated in this line (all for now) */
 899         bitmap_set(line->lun_bitmap, 0, lm->lun_bitmap_len);
 900
 901         smeta->header.identifier = cpu_to_le32(PBLK_MAGIC);
 902         memcpy(smeta->header.uuid, pblk->instance_uuid, 16);
 903         smeta->header.id = cpu_to_le32(line->id);
 904         smeta->header.type = cpu_to_le16(line->type);
 905         smeta->header.version = cpu_to_le16(1);
 906
 907         /* Start metadata */
 908         smeta->seq_nr = cpu_to_le64(line->seq_nr);
 909         smeta->window_wr_lun = cpu_to_le32(geo->nr_luns);
 910
 911         /* Fill metadata among lines */
 912         if (cur) {
 913                 memcpy(line->lun_bitmap, cur->lun_bitmap, lm->lun_bitmap_len);
 914                 smeta->prev_id = cpu_to_le32(cur->id);
 915                 cur->emeta->next_id = cpu_to_le32(line->id);
 916         } else {
 917                 smeta->prev_id = cpu_to_le32(PBLK_LINE_EMPTY);
 918         }
 919
 920         /* All smeta must be set at this point */
 921         smeta->header.crc = cpu_to_le32(pblk_calc_meta_header_crc(pblk, smeta));
 922         smeta->crc = cpu_to_le32(pblk_calc_smeta_crc(pblk, smeta));
 923
 924         /* End metadata */
 925         memcpy(&emeta->header, &smeta->header, sizeof(struct line_header));
 926         emeta->seq_nr = cpu_to_le64(line->seq_nr);
 927         emeta->nr_lbas = cpu_to_le64(line->sec_in_line);
 928         emeta->nr_valid_lbas = cpu_to_le64(0);
 929         emeta->next_id = cpu_to_le32(PBLK_LINE_EMPTY);
 930         emeta->crc = cpu_to_le32(0);
 931         emeta->prev_id = smeta->prev_id;
 932
 933         return 1;
 934 }
 935
 936 /* For now lines are always assumed full lines. Thus, smeta former and current
 937  * lun bitmaps are omitted.
 938  */
 939 static int pblk_line_init_bb(struct pblk *pblk, struct pblk_line *line,
 940                              int init)
 941 {
 942         struct nvm_tgt_dev *dev = pblk->dev;
 943         struct nvm_geo *geo = &dev->geo;
 944         struct pblk_line_meta *lm = &pblk->lm;
 945         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 946         int nr_bb = 0;
 947         u64 off;
 948         int bit = -1;
 949
 950         line->sec_in_line = lm->sec_per_line;
 951
 952         /* Capture bad block information on line mapping bitmaps */
 953         while ((bit = find_next_bit(line->blk_bitmap, lm->blk_per_line,
 954                                         bit + 1)) < lm->blk_per_line) {
 955                 off = bit * geo->sec_per_pl;
 956                 bitmap_shift_left(l_mg->bb_aux, l_mg->bb_template, off,
 957                                                         lm->sec_per_line);
 958                 bitmap_or(line->map_bitmap, line->map_bitmap, l_mg->bb_aux,
 959                                                         lm->sec_per_line);
 960                 line->sec_in_line -= geo->sec_per_blk;
 961                 if (bit >= lm->emeta_bb)
 962                         nr_bb++;
 963         }
 964
 965         /* Mark smeta metadata sectors as bad sectors */
 966         bit = find_first_zero_bit(line->blk_bitmap, lm->blk_per_line);
 967         off = bit * geo->sec_per_pl;
 968 retry_smeta:
 969         bitmap_set(line->map_bitmap, off, lm->smeta_sec);
 970         line->sec_in_line -= lm->smeta_sec;
 971         line->smeta_ssec = off;
 972         line->cur_sec = off + lm->smeta_sec;
 973
 974         if (init && pblk_line_submit_smeta_io(pblk, line, off, WRITE)) {
 975                 pr_debug("pblk: line smeta I/O failed. Retry\n");
 976                 off += geo->sec_per_pl;
 977                 goto retry_smeta;
 978         }
 979
 980         bitmap_copy(line->invalid_bitmap, line->map_bitmap, lm->sec_per_line);
 981
 982         /* Mark emeta metadata sectors as bad sectors. We need to consider bad
 983          * blocks to make sure that there are enough sectors to store emeta
 984          */
 985         bit = lm->sec_per_line;
 986         off = lm->sec_per_line - lm->emeta_sec;
 987         bitmap_set(line->invalid_bitmap, off, lm->emeta_sec);
 988         while (nr_bb) {
 989                 off -= geo->sec_per_pl;
 990                 if (!test_bit(off, line->invalid_bitmap)) {
 991                         bitmap_set(line->invalid_bitmap, off, geo->sec_per_pl);
 992                         nr_bb--;
 993                 }
 994         }
 995
 996         line->sec_in_line -= lm->emeta_sec;
 997         line->emeta_ssec = off;
 998         line->vsc = line->left_ssecs = line->left_msecs = line->sec_in_line;
 999
1000         if (lm->sec_per_line - line->sec_in_line !=
1001                 bitmap_weight(line->invalid_bitmap, lm->sec_per_line)) {
1002                 spin_lock(&line->lock);
1003                 line->state = PBLK_LINESTATE_BAD;
1004                 spin_unlock(&line->lock);
1005
1006                 list_add_tail(&line->list, &l_mg->bad_list);
1007                 pr_err("pblk: unexpected line %d is bad\n", line->id);
1008
1009                 return 0;
1010         }
1011
1012         return 1;
1013 }
1014
1015 static int pblk_line_prepare(struct pblk *pblk, struct pblk_line *line)
1016 {
1017         struct pblk_line_meta *lm = &pblk->lm;
1018         int blk_in_line = atomic_read(&line->blk_in_line);
1019
1020         line->map_bitmap = mempool_alloc(pblk->line_meta_pool, GFP_ATOMIC);
1021         if (!line->map_bitmap)
1022                 return -ENOMEM;
1023         memset(line->map_bitmap, 0, lm->sec_bitmap_len);
1024
1025         /* invalid_bitmap is special since it is used when line is closed. No
1026          * need to zeroized; it will be initialized using bb info form
1027          * map_bitmap
1028          */
1029         line->invalid_bitmap = mempool_alloc(pblk->line_meta_pool, GFP_ATOMIC);
1030         if (!line->invalid_bitmap) {
1031                 mempool_free(line->map_bitmap, pblk->line_meta_pool);
1032                 return -ENOMEM;
1033         }
1034
1035         spin_lock(&line->lock);
1036         if (line->state != PBLK_LINESTATE_FREE) {
1037                 spin_unlock(&line->lock);
1038                 WARN(1, "pblk: corrupted line state\n");
1039                 return -EINTR;
1040         }
1041         line->state = PBLK_LINESTATE_OPEN;
1042
1043         atomic_set(&line->left_eblks, blk_in_line);
1044         atomic_set(&line->left_seblks, blk_in_line);
1045         spin_unlock(&line->lock);
1046
1047         /* Bad blocks do not need to be erased */
1048         bitmap_copy(line->erase_bitmap, line->blk_bitmap, lm->blk_per_line);
1049
1050         kref_init(&line->ref);
1051
1052         return 0;
1053 }
1054
1055 int pblk_line_recov_alloc(struct pblk *pblk, struct pblk_line *line)
1056 {
1057         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1058         int ret;
1059
1060         spin_lock(&l_mg->free_lock);
1061         l_mg->data_line = line;
1062         list_del(&line->list);
1063
1064         ret = pblk_line_prepare(pblk, line);
1065         if (ret) {
1066                 list_add(&line->list, &l_mg->free_list);
1067                 spin_unlock(&l_mg->free_lock);
1068                 return ret;
1069         }
1070         spin_unlock(&l_mg->free_lock);
1071
1072         pblk_rl_free_lines_dec(&pblk->rl, line);
1073
1074         if (!pblk_line_init_bb(pblk, line, 0)) {
1075                 list_add(&line->list, &l_mg->free_list);
1076                 return -EINTR;
1077         }
1078
1079         return 0;
1080 }
1081
1082 void pblk_line_recov_close(struct pblk *pblk, struct pblk_line *line)
1083 {
1084         mempool_free(line->map_bitmap, pblk->line_meta_pool);
1085         line->map_bitmap = NULL;
1086         line->smeta = NULL;
1087         line->emeta = NULL;
1088 }
1089
1090 struct pblk_line *pblk_line_get(struct pblk *pblk)
1091 {
1092         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1093         struct pblk_line_meta *lm = &pblk->lm;
1094         struct pblk_line *line = NULL;
1095         int bit;
1096
1097         lockdep_assert_held(&l_mg->free_lock);
1098
1099 retry_get:
1100         if (list_empty(&l_mg->free_list)) {
1101                 pr_err("pblk: no free lines\n");
1102                 goto out;
1103         }
1104
1105         line = list_first_entry(&l_mg->free_list, struct pblk_line, list);
1106         list_del(&line->list);
1107         l_mg->nr_free_lines--;
1108
1109         bit = find_first_zero_bit(line->blk_bitmap, lm->blk_per_line);
1110         if (unlikely(bit >= lm->blk_per_line)) {
1111                 spin_lock(&line->lock);
1112                 line->state = PBLK_LINESTATE_BAD;
1113                 spin_unlock(&line->lock);
1114
1115                 list_add_tail(&line->list, &l_mg->bad_list);
1116
1117                 pr_debug("pblk: line %d is bad\n", line->id);
1118                 goto retry_get;
1119         }
1120
1121         if (pblk_line_prepare(pblk, line)) {
1122                 pr_err("pblk: failed to prepare line %d\n", line->id);
1123                 list_add(&line->list, &l_mg->free_list);
1124                 return NULL;
1125         }
1126
1127 out:
1128         return line;
1129 }
1130
1131 static struct pblk_line *pblk_line_retry(struct pblk *pblk,
1132                                          struct pblk_line *line)
1133 {
1134         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1135         struct pblk_line *retry_line;
1136
1137         spin_lock(&l_mg->free_lock);
1138         retry_line = pblk_line_get(pblk);
1139         if (!retry_line) {
1140                 l_mg->data_line = NULL;
1141                 spin_unlock(&l_mg->free_lock);
1142                 return NULL;
1143         }
1144
1145         retry_line->smeta = line->smeta;
1146         retry_line->emeta = line->emeta;
1147         retry_line->meta_line = line->meta_line;
1148
1149         pblk_line_free(pblk, line);
1150         l_mg->data_line = retry_line;
1151         spin_unlock(&l_mg->free_lock);
1152
1153         if (pblk_line_erase(pblk, retry_line)) {
1154                 spin_lock(&l_mg->free_lock);
1155                 l_mg->data_line = NULL;
1156                 spin_unlock(&l_mg->free_lock);
1157                 return NULL;
1158         }
1159
1160         pblk_rl_free_lines_dec(&pblk->rl, retry_line);
1161
1162         return retry_line;
1163 }
1164
1165 struct pblk_line *pblk_line_get_first_data(struct pblk *pblk)
1166 {
1167         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1168         struct pblk_line *line;
1169         int meta_line;
1170         int is_next = 0;
1171
1172         spin_lock(&l_mg->free_lock);
1173         line = pblk_line_get(pblk);
1174         if (!line) {
1175                 spin_unlock(&l_mg->free_lock);
1176                 return NULL;
1177         }
1178
1179         line->seq_nr = l_mg->d_seq_nr++;
1180         line->type = PBLK_LINETYPE_DATA;
1181         l_mg->data_line = line;
1182
1183         meta_line = find_first_zero_bit(&l_mg->meta_bitmap, PBLK_DATA_LINES);
1184         set_bit(meta_line, &l_mg->meta_bitmap);
1185         line->smeta = l_mg->sline_meta[meta_line].meta;
1186         line->emeta = l_mg->eline_meta[meta_line].meta;
1187         line->meta_line = meta_line;
1188
1189         /* Allocate next line for preparation */
1190         l_mg->data_next = pblk_line_get(pblk);
1191         if (l_mg->data_next) {
1192                 l_mg->data_next->seq_nr = l_mg->d_seq_nr++;
1193                 l_mg->data_next->type = PBLK_LINETYPE_DATA;
1194                 is_next = 1;
1195         }
1196         spin_unlock(&l_mg->free_lock);
1197
1198         pblk_rl_free_lines_dec(&pblk->rl, line);
1199         if (is_next)
1200                 pblk_rl_free_lines_dec(&pblk->rl, l_mg->data_next);
1201
1202         if (pblk_line_erase(pblk, line))
1203                 return NULL;
1204
1205 retry_setup:
1206         if (!pblk_line_set_metadata(pblk, line, NULL)) {
1207                 line = pblk_line_retry(pblk, line);
1208                 if (!line)
1209                         return NULL;
1210
1211                 goto retry_setup;
1212         }
1213
1214         if (!pblk_line_init_bb(pblk, line, 1)) {
1215                 line = pblk_line_retry(pblk, line);
1216                 if (!line)
1217                         return NULL;
1218
1219                 goto retry_setup;
1220         }
1221
1222         return line;
1223 }
1224
1225 struct pblk_line *pblk_line_replace_data(struct pblk *pblk)
1226 {
1227         struct pblk_line_meta *lm = &pblk->lm;
1228         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1229         struct pblk_line *cur, *new;
1230         unsigned int left_seblks;
1231         int meta_line;
1232         int is_next = 0;
1233
1234         cur = l_mg->data_line;
1235         new = l_mg->data_next;
1236         if (!new)
1237                 return NULL;
1238         l_mg->data_line = new;
1239
1240 retry_line:
1241         left_seblks = atomic_read(&new->left_seblks);
1242         if (left_seblks) {
1243                 /* If line is not fully erased, erase it */
1244                 if (atomic_read(&new->left_eblks)) {
1245                         if (pblk_line_erase(pblk, new))
1246                                 return NULL;
1247                 } else {
1248                         io_schedule();
1249                 }
1250                 goto retry_line;
1251         }
1252
1253         spin_lock(&l_mg->free_lock);
1254         /* Allocate next line for preparation */
1255         l_mg->data_next = pblk_line_get(pblk);
1256         if (l_mg->data_next) {
1257                 l_mg->data_next->seq_nr = l_mg->d_seq_nr++;
1258                 l_mg->data_next->type = PBLK_LINETYPE_DATA;
1259                 is_next = 1;
1260         }
1261
1262 retry_meta:
1263         meta_line = find_first_zero_bit(&l_mg->meta_bitmap, PBLK_DATA_LINES);
1264         if (meta_line == PBLK_DATA_LINES) {
1265                 spin_unlock(&l_mg->free_lock);
1266                 io_schedule();
1267                 spin_lock(&l_mg->free_lock);
1268                 goto retry_meta;
1269         }
1270
1271         set_bit(meta_line, &l_mg->meta_bitmap);
1272         new->smeta = l_mg->sline_meta[meta_line].meta;
1273         new->emeta = l_mg->eline_meta[meta_line].meta;
1274         new->meta_line = meta_line;
1275
1276         memset(new->smeta, 0, lm->smeta_len);
1277         memset(new->emeta, 0, lm->emeta_len);
1278         spin_unlock(&l_mg->free_lock);
1279
1280         if (is_next)
1281                 pblk_rl_free_lines_dec(&pblk->rl, l_mg->data_next);
1282
1283 retry_setup:
1284         if (!pblk_line_set_metadata(pblk, new, cur)) {
1285                 new = pblk_line_retry(pblk, new);
1286                 if (!new)
1287                         return NULL;
1288
1289                 goto retry_setup;
1290         }
1291
1292         if (!pblk_line_init_bb(pblk, new, 1)) {
1293                 new = pblk_line_retry(pblk, new);
1294                 if (!new)
1295                         return NULL;
1296
1297                 goto retry_setup;
1298         }
1299
1300         return new;
1301 }
1302
1303 void pblk_line_free(struct pblk *pblk, struct pblk_line *line)
1304 {
1305         if (line->map_bitmap)
1306                 mempool_free(line->map_bitmap, pblk->line_meta_pool);
1307         if (line->invalid_bitmap)
1308                 mempool_free(line->invalid_bitmap, pblk->line_meta_pool);
1309
1310         line->map_bitmap = NULL;
1311         line->invalid_bitmap = NULL;
1312         line->smeta = NULL;
1313         line->emeta = NULL;
1314 }
1315
1316 void pblk_line_put(struct kref *ref)
1317 {
1318         struct pblk_line *line = container_of(ref, struct pblk_line, ref);
1319         struct pblk *pblk = line->pblk;
1320         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1321
1322         spin_lock(&line->lock);
1323         WARN_ON(line->state != PBLK_LINESTATE_GC);
1324         line->state = PBLK_LINESTATE_FREE;
1325         line->gc_group = PBLK_LINEGC_NONE;
1326         pblk_line_free(pblk, line);
1327         spin_unlock(&line->lock);
1328
1329         spin_lock(&l_mg->free_lock);
1330         list_add_tail(&line->list, &l_mg->free_list);
1331         l_mg->nr_free_lines++;
1332         spin_unlock(&l_mg->free_lock);
1333
1334         pblk_rl_free_lines_inc(&pblk->rl, line);
1335 }
1336
1337 int pblk_blk_erase_async(struct pblk *pblk, struct ppa_addr ppa)
1338 {
1339         struct nvm_rq *rqd;
1340         int err;
1341
1342         rqd = mempool_alloc(pblk->r_rq_pool, GFP_KERNEL);
1343         memset(rqd, 0, pblk_r_rq_size);
1344
1345         pblk_setup_e_rq(pblk, rqd, ppa);
1346
1347         rqd->end_io = pblk_end_io_erase;
1348         rqd->private = pblk;
1349
1350         /* The write thread schedules erases so that it minimizes disturbances
1351          * with writes. Thus, there is no need to take the LUN semaphore.
1352          */
1353         err = pblk_submit_io(pblk, rqd);
1354         if (err) {
1355                 struct nvm_tgt_dev *dev = pblk->dev;
1356                 struct nvm_geo *geo = &dev->geo;
1357
1358                 pr_err("pblk: could not async erase line:%d,blk:%d\n",
1359                                         pblk_dev_ppa_to_line(ppa),
1360                                         pblk_dev_ppa_to_pos(geo, ppa));
1361         }
1362
1363         return err;
1364 }
1365
1366 struct pblk_line *pblk_line_get_data(struct pblk *pblk)
1367 {
1368         return pblk->l_mg.data_line;
1369 }
1370
1371 struct pblk_line *pblk_line_get_data_next(struct pblk *pblk)
1372 {
1373         return pblk->l_mg.data_next;
1374 }
1375
1376 int pblk_line_is_full(struct pblk_line *line)
1377 {
1378         return (line->left_msecs == 0);
1379 }
1380
1381 void pblk_line_close(struct pblk *pblk, struct pblk_line *line)
1382 {
1383         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1384         struct list_head *move_list;
1385
1386         line->emeta->crc = cpu_to_le32(pblk_calc_emeta_crc(pblk, line->emeta));
1387
1388         if (pblk_line_submit_emeta_io(pblk, line, line->cur_sec, WRITE))
1389                 pr_err("pblk: line %d close I/O failed\n", line->id);
1390
1391         WARN(!bitmap_full(line->map_bitmap, line->sec_in_line),
1392                                 "pblk: corrupt closed line %d\n", line->id);
1393
1394         spin_lock(&l_mg->free_lock);
1395         WARN_ON(!test_and_clear_bit(line->meta_line, &l_mg->meta_bitmap));
1396         spin_unlock(&l_mg->free_lock);
1397
1398         spin_lock(&l_mg->gc_lock);
1399         spin_lock(&line->lock);
1400         WARN_ON(line->state != PBLK_LINESTATE_OPEN);
1401         line->state = PBLK_LINESTATE_CLOSED;
1402         move_list = pblk_line_gc_list(pblk, line);
1403
1404         list_add_tail(&line->list, move_list);
1405
1406         mempool_free(line->map_bitmap, pblk->line_meta_pool);
1407         line->map_bitmap = NULL;
1408         line->smeta = NULL;
1409         line->emeta = NULL;
1410
1411         spin_unlock(&line->lock);
1412         spin_unlock(&l_mg->gc_lock);
1413 }
1414
1415 void pblk_line_close_ws(struct work_struct *work)
1416 {
1417         struct pblk_line_ws *line_ws = container_of(work, struct pblk_line_ws,
1418                                                                         ws);
1419         struct pblk *pblk = line_ws->pblk;
1420         struct pblk_line *line = line_ws->line;
1421
1422         pblk_line_close(pblk, line);
1423         mempool_free(line_ws, pblk->line_ws_pool);
1424 }
1425
1426 void pblk_line_mark_bb(struct work_struct *work)
1427 {
1428         struct pblk_line_ws *line_ws = container_of(work, struct pblk_line_ws,
1429                                                                         ws);
1430         struct pblk *pblk = line_ws->pblk;
1431         struct nvm_tgt_dev *dev = pblk->dev;
1432         struct ppa_addr *ppa = line_ws->priv;
1433         int ret;
1434
1435         ret = nvm_set_tgt_bb_tbl(dev, ppa, 1, NVM_BLK_T_GRWN_BAD);
1436         if (ret) {
1437                 struct pblk_line *line;
1438                 int pos;
1439
1440                 line = &pblk->lines[pblk_dev_ppa_to_line(*ppa)];
1441                 pos = pblk_dev_ppa_to_pos(&dev->geo, *ppa);
1442
1443                 pr_err("pblk: failed to mark bb, line:%d, pos:%d\n",
1444                                 line->id, pos);
1445         }
1446
1447         kfree(ppa);
1448         mempool_free(line_ws, pblk->line_ws_pool);
1449 }
1450
1451 void pblk_line_run_ws(struct pblk *pblk, struct pblk_line *line, void *priv,
1452                       void (*work)(struct work_struct *))
1453 {
1454         struct pblk_line_ws *line_ws;
1455
1456         line_ws = mempool_alloc(pblk->line_ws_pool, GFP_ATOMIC);
1457         if (!line_ws)
1458                 return;
1459
1460         line_ws->pblk = pblk;
1461         line_ws->line = line;
1462         line_ws->priv = priv;
1463
1464         INIT_WORK(&line_ws->ws, work);
1465         queue_work(pblk->kw_wq, &line_ws->ws);
1466 }
1467
1468 void pblk_down_rq(struct pblk *pblk, struct ppa_addr *ppa_list, int nr_ppas,
1469                   unsigned long *lun_bitmap)
1470 {
1471         struct nvm_tgt_dev *dev = pblk->dev;
1472         struct nvm_geo *geo = &dev->geo;
1473         struct pblk_lun *rlun;
1474         int lun_id = ppa_list[0].g.ch * geo->luns_per_chnl + ppa_list[0].g.lun;
1475         int ret;
1476
1477         /*
1478          * Only send one inflight I/O per LUN. Since we map at a page
1479          * granurality, all ppas in the I/O will map to the same LUN
1480          */
1481 #ifdef CONFIG_NVM_DEBUG
1482         int i;
1483
1484         for (i = 1; i < nr_ppas; i++)
1485                 WARN_ON(ppa_list[0].g.lun != ppa_list[i].g.lun ||
1486                                 ppa_list[0].g.ch != ppa_list[i].g.ch);
1487 #endif
1488         /* If the LUN has been locked for this same request, do no attempt to
1489          * lock it again
1490          */
1491         if (test_and_set_bit(lun_id, lun_bitmap))
1492                 return;
1493
1494         rlun = &pblk->luns[lun_id];
1495         ret = down_timeout(&rlun->wr_sem, msecs_to_jiffies(5000));
1496         if (ret) {
1497                 switch (ret) {
1498                 case -ETIME:
1499                         pr_err("pblk: lun semaphore timed out\n");
1500                         break;
1501                 case -EINTR:
1502                         pr_err("pblk: lun semaphore timed out\n");
1503                         break;
1504                 }
1505         }
1506 }
1507
1508 void pblk_up_rq(struct pblk *pblk, struct ppa_addr *ppa_list, int nr_ppas,
1509                 unsigned long *lun_bitmap)
1510 {
1511         struct nvm_tgt_dev *dev = pblk->dev;
1512         struct nvm_geo *geo = &dev->geo;
1513         struct pblk_lun *rlun;
1514         int nr_luns = geo->nr_luns;
1515         int bit = -1;
1516
1517         while ((bit = find_next_bit(lun_bitmap, nr_luns, bit + 1)) < nr_luns) {
1518                 rlun = &pblk->luns[bit];
1519                 up(&rlun->wr_sem);
1520         }
1521
1522         kfree(lun_bitmap);
1523 }
1524
1525 void pblk_update_map(struct pblk *pblk, sector_t lba, struct ppa_addr ppa)
1526 {
1527         struct ppa_addr l2p_ppa;
1528
1529         /* logic error: lba out-of-bounds. Ignore update */
1530         if (!(lba < pblk->rl.nr_secs)) {
1531                 WARN(1, "pblk: corrupted L2P map request\n");
1532                 return;
1533         }
1534
1535         spin_lock(&pblk->trans_lock);
1536         l2p_ppa = pblk_trans_map_get(pblk, lba);
1537
1538         if (!pblk_addr_in_cache(l2p_ppa) && !pblk_ppa_empty(l2p_ppa))
1539                 pblk_map_invalidate(pblk, l2p_ppa);
1540
1541         pblk_trans_map_set(pblk, lba, ppa);
1542         spin_unlock(&pblk->trans_lock);
1543 }
1544
1545 void pblk_update_map_cache(struct pblk *pblk, sector_t lba, struct ppa_addr ppa)
1546 {
1547 #ifdef CONFIG_NVM_DEBUG
1548         /* Callers must ensure that the ppa points to a cache address */
1549         BUG_ON(!pblk_addr_in_cache(ppa));
1550         BUG_ON(pblk_rb_pos_oob(&pblk->rwb, pblk_addr_to_cacheline(ppa)));
1551 #endif
1552
1553         pblk_update_map(pblk, lba, ppa);
1554 }
1555
1556 int pblk_update_map_gc(struct pblk *pblk, sector_t lba, struct ppa_addr ppa,
1557                        struct pblk_line *gc_line)
1558 {
1559         struct ppa_addr l2p_ppa;
1560         int ret = 1;
1561
1562 #ifdef CONFIG_NVM_DEBUG
1563         /* Callers must ensure that the ppa points to a cache address */
1564         BUG_ON(!pblk_addr_in_cache(ppa));
1565         BUG_ON(pblk_rb_pos_oob(&pblk->rwb, pblk_addr_to_cacheline(ppa)));
1566 #endif
1567
1568         /* logic error: lba out-of-bounds. Ignore update */
1569         if (!(lba < pblk->rl.nr_secs)) {
1570                 WARN(1, "pblk: corrupted L2P map request\n");
1571                 return 0;
1572         }
1573
1574         spin_lock(&pblk->trans_lock);
1575         l2p_ppa = pblk_trans_map_get(pblk, lba);
1576
1577         /* Prevent updated entries to be overwritten by GC */
1578         if (pblk_addr_in_cache(l2p_ppa) || pblk_ppa_empty(l2p_ppa) ||
1579                                 pblk_tgt_ppa_to_line(l2p_ppa) != gc_line->id) {
1580                 ret = 0;
1581                 goto out;
1582         }
1583
1584         pblk_trans_map_set(pblk, lba, ppa);
1585 out:
1586         spin_unlock(&pblk->trans_lock);
1587         return ret;
1588 }
1589
1590 void pblk_update_map_dev(struct pblk *pblk, sector_t lba, struct ppa_addr ppa,
1591                          struct ppa_addr entry_line)
1592 {
1593         struct ppa_addr l2p_line;
1594
1595 #ifdef CONFIG_NVM_DEBUG
1596         /* Callers must ensure that the ppa points to a device address */
1597         BUG_ON(pblk_addr_in_cache(ppa));
1598 #endif
1599         /* Invalidate and discard padded entries */
1600         if (lba == ADDR_EMPTY) {
1601 #ifdef CONFIG_NVM_DEBUG
1602                 atomic_long_inc(&pblk->padded_wb);
1603 #endif
1604                 pblk_map_invalidate(pblk, ppa);
1605                 return;
1606         }
1607
1608         /* logic error: lba out-of-bounds. Ignore update */
1609         if (!(lba < pblk->rl.nr_secs)) {
1610                 WARN(1, "pblk: corrupted L2P map request\n");
1611                 return;
1612         }
1613
1614         spin_lock(&pblk->trans_lock);
1615         l2p_line = pblk_trans_map_get(pblk, lba);
1616
1617         /* Do not update L2P if the cacheline has been updated. In this case,
1618          * the mapped ppa must be invalidated
1619          */
1620         if (l2p_line.ppa != entry_line.ppa) {
1621                 if (!pblk_ppa_empty(ppa))
1622                         pblk_map_invalidate(pblk, ppa);
1623                 goto out;
1624         }
1625
1626 #ifdef CONFIG_NVM_DEBUG
1627         WARN_ON(!pblk_addr_in_cache(l2p_line) && !pblk_ppa_empty(l2p_line));
1628 #endif
1629
1630         pblk_trans_map_set(pblk, lba, ppa);
1631 out:
1632         spin_unlock(&pblk->trans_lock);
1633 }
1634
1635 void pblk_lookup_l2p_seq(struct pblk *pblk, struct ppa_addr *ppas,
1636                          sector_t blba, int nr_secs)
1637 {
1638         int i;
1639
1640         spin_lock(&pblk->trans_lock);
1641         for (i = 0; i < nr_secs; i++)
1642                 ppas[i] = pblk_trans_map_get(pblk, blba + i);
1643         spin_unlock(&pblk->trans_lock);
1644 }
1645
1646 void pblk_lookup_l2p_rand(struct pblk *pblk, struct ppa_addr *ppas,
1647                           u64 *lba_list, int nr_secs)
1648 {
1649         sector_t lba;
1650         int i;
1651
1652         spin_lock(&pblk->trans_lock);
1653         for (i = 0; i < nr_secs; i++) {
1654                 lba = lba_list[i];
1655                 if (lba == ADDR_EMPTY) {
1656                         ppas[i].ppa = ADDR_EMPTY;
1657                 } else {
1658                         /* logic error: lba out-of-bounds. Ignore update */
1659                         if (!(lba < pblk->rl.nr_secs)) {
1660                                 WARN(1, "pblk: corrupted L2P map request\n");
1661                                 continue;
1662                         }
1663                         ppas[i] = pblk_trans_map_get(pblk, lba);
1664                 }
1665         }
1666         spin_unlock(&pblk->trans_lock);
1667 }